The Intelligence Explosion refers to a concept
invented by the statistician, I. J. Good. Good was
interested in Bayesian statistics, and learned to play go
from Alan Turing. In his own words:
Let an ultraintelligent machine be defined as a machine
that can far surpass all the intellectual activities of any
man however clever. Since the design of machines is one of
these intellectual activities, an ultraintelligent machine
could design even better machines; there would then
unquestionably be an "intelligence explosion," and the
intelligence of man would be left far behind [...]. Thus the
first ultraintelligent machine is the last invention that
man need ever make, provided that the machine is docile
enough to tell us how to keep it under control.
- I. J. Good (1965)
Good's concept has been used widely in modern times to
support the idea that - at some point in the future,
intelligent self-modifying machines will arise, and
then an "Intelligence Explosion" will occur.
Nick Bostrom expresses the idea as follows:
"Emergence of superintelligence may be sudden.
It appears much harder to get from where we are now to
human-level artificial intelligence than to get from there
to superintelligence. While it may thus take quite a while
before we get superintelligence, the final stage may happen
swiftly. That is, the transition from a state where we have
a roughly human-level artificial intelligence to a state
where we have full-blown superintelligence, with
revolutionary applications, may be very rapid, perhaps a
matter of days rather than years. This possibility of a
sudden emergence of superintelligence is referred to as the
From the standpoint of existential risk, one of the most
critical points about Artificial Intelligence is that an
Artificial Intelligence might increase in intelligence
extremely fast. The obvious reason to suspect this
possibility is recursive self-improvement. (Good 1965.) The
AI becomes smarter, including becoming smarter at the task
of writing the internal cognitive functions of an AI, so
the AI can rewrite its existing cognitive functions to work
even better, which makes the AI still smarter, including
smarter at the task of rewriting itself, so that it makes
yet more improvements. [...] The key implication for our
purposes is that an AI might make a huge jump in
intelligence after reaching some threshold of criticality.
These ideas lead to the question of when such a
scenario is going to start happening.
The Intelligence Explosion Is Happening Now
This essay expounds on the thesis that the intelligence
explosion is already happening.
How can that possibly be - when Good defined
his intelligence explosion as being the result of the
existence of an ultraintelligent machine - and there
are currently none of those in sight?
The answer is that Good did not formulate his concept
in such a way that it "carves nature at the joints" -
and that - properly understood - the real underlying
concept - to which the term "intelligence explosion"
should properly refer - describes a process which is
An intelligence explosion of the type which Good
described does not logically require an
"ultraintelligent machine" (i.e. a machine that can far
surpass all the intellectual activities of any man
however clever). All it needs is an intelligent system
which is able to modify its own algorithm.
We already have an intelligent system which
modifies itself. It is known as "the man-machine
Machines are already heavily involved in the
design of other machines. No-one could design a modern CPU
without the use of computers. No one could build one without
the help of sophisticated machinery.
The idea that machines will suddenly take over
this task when they become smart enough seems naive.
Rather there is a man-machine symbiosis involved the design
of machine - with the "man" part being gradually replaced by
Machines do not yet make expert computer programmers - but
they already contribute massively to computer programming
Refactoring: Refactoring involves performing
rearrangements of code which preserve its function, and improve its
readability and maintainability - or facilitate future improvements.
Much refactoring is done by daemons - and their existence
massively speeds up the production of working code. Refactoring
daemons enable tasks which would previously have been intractable.
Specification languages: High level languages are
another example of programming by machines. When the author started
programming, everything was done in machine code. With 32K of RAM -
most of which was devoted to screen memory - optimisation, compression
and memory management occupied a lot of programming time. These days,
most of the donkey work of programming is performed by
mechanically - humans get to work in a cushy high-level environment
where they don't have to bother themselves with tasks like
collecting garbage, or deciding in advance how big their data
structures are. Ultimately, specifications will be
translated into code in this way.
Machines also automatically detect programming
errors, and automatically test existing programs.
Machines will gradually get better at these kinds of computer
programming tasks - taking over an increasing quantity of the load
from humans. This includes tasks that involve modifying their own
The man-machine symbiosis
That the intelligence explosion can be applied to the man-machine
symbiosis has been recognised before - notably by Eliezer Yudkowsky:
I. J. Good only talked about AI, but in principle the concept of an
intelligence explosion generalizes further: for example, humans
augmented by direct brain-computer interfaces, using their improved
intelligence to design better brain-computer interfaces.
Of course, we have had brain-computer interfaces for decades. They
involve keyboards, mice, eyes, arms, hands and software.
Won't there be a sudden speed-up when sluggish humans are
finally eliminated from the loop? Probably not. By that point,
machine "code wizards" will be writing most of the code
anyway - so progress will already be pretty rapid. However,
humans will most likely want to keep an eye on their budding machine
programmers for a while - so there will be regular code reviews. As
confidence is gained in the results, the reviews will be needed less
frequently. So: humans will probably not drop out suddenly - but
rather gradually, with increasingly-infrequent compulsory
It is a fallacy to argue that today's machines are designed by humans,
and - since the intelligence of individual humans is not increasing -
the intelligence explosion has not started yet. Today's machines are
actually designed by networks of humans with the
help of machines. The machines are currently
improving - and so are the networking technologies that link humans
with other humans and with machines. Machines pre-process our sensory
data, and post-process our motor outputs - and the results are smarter
than we are alone.
So: machine intelligence will not suddenly surpass humans.
Rather, machines have been gradually beating humans - an
application domain at a time - for decades now. A machine
intelligence that is of "roughly human-level" is actually
likely to be either vastly superior in some domans or vastly inferior
in others - simply because machine intelligence so far has proven to
be so vastly different from our own in terms of its strengths and
weaknesses - so machines will take over tasks from humans gradually,
a domain at a time as they increase in comptence.
These considerations suggest that there will be no terribly sudden
"intelligence explosion" which starts at some point in the future -
rather, the intelligence explosion is happening now.
The origins of the intelligence explosion
One of Robin Hanson's slides illustrates exponential growth in brain
size dating back to the origin of animals:
This graph is based on data from Harry Jerison's book: "Brain Size and the Evolution of the Mind."
The largest nervous systems doubled in size about
every fifteen million years since the Cambrian explosion 550 million
years ago. Robot controllers double in complexity (processing power)
every year or two.
The intelligence explosion can also be traced back through our hominid
In modern times, intelligence has been augmented by networking
technologies, that enable minds to be linked together, mind-machine
interfaces, and the internet. Both humans and companies have increased
in their intellectual abilities dramatically as a result.
The origin of the intelligence explosion can be traced back
to when evolution first developed simulation technology - in the form
of brains, and began to perform evaluations under simulation. It
accelerated when organisms began to select mates using intelligence.
Use of non-nucleic inheritance accelerated the explosion further, as
did tool use, networking and the internet.
Intelligent self improving systems
A kind of corollory of the point made in this essay is
that there already exists an intelligent self improving
system - namely the man-machine civilisation.
I am not sure that this point is widely understood
In the "Self-Improving Artificial Intelligence" talk
listed in the reference section, Eric Drexler asks
after an existence proof of an intelligent
self-improving system. It does seem reasonable
to me to point to the man-machine civilisation as an
existence proof of an intelligent self-improving system -
God's Utility Function.
The technology explosion
We are in the midst of the "intelligence explosion" now. It
is part of the ongoing
"technology explosion", which
affects many aspects of technology - not just
The "technology explosion" is really just another name for evolution -
with pioneering adaptations being seen as a kind of natural
technology - put together by tinkering, rather than by engineers.
The technology explosion started long ago - and will likely
reverberate into the future for some time to come.